Comparison of PCA and SVM for a West Indian Script-Gujarati

Through the dawn of technical era, translation of scanned document (handwritten or printed) into machine editable format has attracted many researchers. Gujarati is spoken and used as official language in Gujarat, a western state in India. In this paper an attempt is made to compare the offline recognition system for the isolated handwritten Gujarati numerals with database size of 800 numerals. As database was not available it has been created by us by taking samples from different people on specially designed sheet. For feature extraction affine invariant moments based model is used. We are using SVM classifier and PCA (to reduce dimensions of feature space) and used Euclidean similarity measure to classify the numerals. SVM classifier yielded 92% as recognition rate whereas PCA scored recognition rate of 84%. The comparison of SVM and PCA is made and it can be seen that SVM classifier has shown better results as compared to PCA classifier.